OzBrain vs ChatGPT Projects
Updated
The short answer
ChatGPT Projects are containers inside ChatGPT: files, instructions, and a project memory shared by the chats in one workspace, free on every plan. OzBrain is the brain underneath: one set of structured, checked articles that every project, and every agent beyond ChatGPT, reads and writes. Projects organize your chats. The brain holds what all of them, and all of you, know.
If your work lives in tidy project boxes already, the organizing is working. What leaks is the knowledge: out of old chats, out of pruned memory, and never across the walls between boxes.
Free plan includes 50 articles. One connector approval in ChatGPT.
Side by side
| Dimension | OzBrain | ChatGPT Projects |
|---|---|---|
| Scope | Under every box, behind every agent | One workspace in ChatGPT |
| Cross-chat memoryProject memory keeps snippets and prunes them as space fills. | Persists until superseded | Pruned snippets |
| FilesGenerated files do not save back as project knowledge. | Written back as agents work | Re-uploaded by hand |
| What you can audit | Every article and version | No visible list |
| ReachShared projects stay inside ChatGPT. A brain grant reaches other agents and other people’s agents. | Every agent, plus people you grant | ChatGPT only |
Where does what you learned actually go?
Project memory is selective snippets, not a record. The system keeps what it judges important, older logs get replaced as the space fills, there is no list you can audit, and in practice it misses details from sibling chats. Files a project generates do not even save back as project knowledge; you re-upload your own outputs by hand.
The walls are load-bearing, and they cut both ways. Files cannot be shared across projects, so the same reference gets uploaded into each box and every copy ages separately. Shared projects can invite collaborators inside ChatGPT; nothing inside a project reaches Claude, your editor, or the project next door, and a teammate’s agents outside ChatGPT still cannot read it.
OzBrain sits underneath instead of inside. Agents stage what matters as they work, the brain routes each write to its article and checks it against what it already holds, and every chat in every project reads the same current version. So do your other agents, and so do the people you share the brain with: a grant puts your shared knowledge behind their agents too, with every write attributed to its author.
The layer under every project box takes one sign-in to create.
Put a brain under my projectsThe boxes earn their keep
Files, instructions, and chats for one effort live in one place, every chat in the project reads the same uploads, and chats inside a project can reference each other. Project-only memory is a real boundary too: client work that must stay contained, stays contained. All of it costs nothing on any plan.
Connect once, and every box reads the brain
Add OzBrain from the connector menu in ChatGPT settings and approve it. Any chat, inside a project or out, can then read the routing index and stage writes. Ask ChatGPT to move what your projects have settled into the brain: the shared references, the decisions, the facts you keep re-uploading.
Per-project uploads shrink to what is genuinely specific to that effort, and the copies that used to age separately become one article with one history.
Keep the boxes. Put a brain under them.
Projects keep scoping the work and containing what must stay contained. The brain carries what should outlive any box: identical in every project, current in every agent, and readable by the collaborators you let in.
Connect it now and the next thing a project chat learns is the last thing you ever re-explain to a sibling chat. Sharing ships with Pro and Max.
Common questions
Do Projects and OzBrain stack?
Yes. A project scopes the effort; the brain supplies the knowledge. Connect OzBrain once and every chat reads the same brain, so per-project uploads shrink to what is truly specific to that effort.
Is project-only memory not enough isolation?
It is real isolation, and it works. It is also the point: every isolated box holds another partial version of you. The brain is the layer under the boxes, structured so an agent takes what the task needs instead of inheriting a wall.
Can my team share one brain across their ChatGPT accounts?
Yes. Grant each account read or read and write access and their ChatGPT, along with every other agent they use, reads the shared brain. Writes resolve through the staged write protocol with provenance per author. Sharing ships with Pro and Max.
What about Claude Projects?
Same shape, harder walls: chats in a Claude project do not share context at all unless it is in the knowledge base. The full comparison is at OzBrain vs Claude Projects.